Multiple memory systems as substrates for multiple decision systems | JoVE Visualize

Area of Science:

Cognitive Neuroscience
Computational Neuroscience
Decision Science

Background:

Value-based decision making involves multiple computational strategies, including habitual (model-free reinforcement learning) and goal-directed (model-based reinforcement learning) actions.
A parallel dissociation exists between memory systems: procedural learning (akin to model-free RL) and declarative memory (supporting relational encoding via the hippocampus).
The hippocampus's role in relational encoding is crucial for learning internal models central to model-based RL.

Purpose of the Study:

To investigate the relationship between model-based reinforcement learning and relational memory.
To determine if individual differences in relational memory capacity correlate with the use of model-based versus model-free strategies in decision-making tasks.

Main Methods:

Human subjects performed two tasks: acquired equivalence (measuring relational encoding and hippocampal-dependent memory) and a sequential reinforcement learning (RL) task.
Individual differences in generalization within the acquired equivalence task were assessed.
The correlation between generalization performance and strategy selection (model-based vs. model-free RL) in the decision task was analyzed.

Main Results:

A significant positive relationship was observed between generalization in the acquired equivalence task and the use of model-based choice strategies in the RL task.
No significant relationship was found between generalization and model-free choice strategies.
These findings support the hypothesis that model-based RL relies on a general-purpose relational memory system.

Conclusions:

Model-based reinforcement learning and relational memory, as measured by acquired equivalence, share underlying cognitive mechanisms.
The results suggest that the capacity for relational memory supports the flexible, goal-directed computations characteristic of model-based decision making.
This provides evidence for a grounding of decision-making system dissociations within memory system dissociations.